barrier

Hi, I am new to OpenCL.

There is a problem making me confused.

Without barrier, C[0] and C[1] have the different value.

C[0] and C[1] should get the same value with the barrier function.

the code get the wrong answer.

where is the bug in the code?

opencl code

__kernel void barrier_example ( __global int *C )

{

//Get the index of the current element

    int t = get_local_id(0);   

    __local int *a1 ;   

    a1 = 0 ;

//barrier test

    if ( t == 1) {    

       for( int j = 0 ; j < 1000 ; j ++  ) ;

   a1 = 100 ;       

    }

barrier ( CLK_LOCAL_MEM_FENCE ) ;

if ( t == 0 ) C [ t ] = a1 ;   

    if ( t == 1 ) C [ t ] = a1 ;

    C [ 2 ] = 88 ;

}

main.c

#include <stdio.h>

#include <stdlib.h>

#ifdef __APPLE__

#include <OpenCL/opencl.h>

#else

#include <CL/cl.h>

#endif

#define MAX_SOURCE_SIZE (0x100000)

int main(void) {

    // Create the two input vectors   

    const int LIST_SIZE = 1024;

// Load the kernel source code into the array source_str

    FILE *fp;

    char *source_str;

    size_t source_size;

fp = fopen("barrier_example.cl", "r");

    if (!fp) {

        fprintf(stderr, "Failed to load kernel.\n");

        exit(1);

    }

    source_str = (char*)malloc(MAX_SOURCE_SIZE);

    source_size = fread( source_str, 1, MAX_SOURCE_SIZE, fp);

    fclose( fp );

// Get platform and device information

    cl_platform_id platform_id = NULL;

    cl_device_id device_id = NULL;   

    cl_uint ret_num_devices;

    cl_uint ret_num_platforms;

    cl_int ret = clGetPlatformIDs(1, &platform_id, &ret_num_platforms);

    ret = clGetDeviceIDs( platform_id, CL_DEVICE_TYPE_ALL, 1, 

            &device_id, &ret_num_devices);

// Create an OpenCL context

    cl_context context = clCreateContext( NULL, 1, &device_id, NULL, NULL, &ret);

// Create a command queue

    cl_command_queue command_queue = clCreateCommandQueue(context, device_id, 0, &ret);

// Create memory buffers on the device for each vector 

    cl_mem c_mem_obj = clCreateBuffer(context, CL_MEM_WRITE_ONLY, 

            LIST_SIZE * sizeof(int), NULL, &ret);

// Create a program from the kernel source

    cl_program program = clCreateProgramWithSource(context, 1, 

            (const char **)&source_str, (const size_t *)&source_size, &ret);

// Build the program

    ret = clBuildProgram(program, 1, &device_id, NULL, NULL, NULL);

// Create the OpenCL kernel

    cl_kernel kernel = clCreateKernel(program, "barrier_example", &ret);

// Set the arguments of the kernel

    ret = clSetKernelArg(kernel, 0, sizeof(cl_mem), (void *)&c_mem_obj);

// Execute the OpenCL kernel on the list

    size_t global_item_size = 6 ; // Process the entire lists

    size_t local_item_size = 3 ; // Process in groups of 64

    ret = clEnqueueNDRangeKernel(command_queue, kernel, 1, NULL, 

            &global_item_size, &local_item_size, 0, NULL, NULL);

// Read the memory buffer C on the device to the local variable C

    int *C = (int*)malloc(sizeof(int)*3*2);

    ret = clEnqueueReadBuffer(command_queue, c_mem_obj, CL_TRUE, 0, 

            3*2 * sizeof(int), C, 0, NULL, NULL);

// Display the result to the screen

    for(int i = 0; i < 3; i++)

        printf("i= %d , %d\n",  i, C[i]);

// Clean up

    ret = clFlush(command_queue);

    ret = clFinish(command_queue);

    ret = clReleaseKernel(kernel);

    ret = clReleaseProgram(program);

ret = clReleaseMemObject(c_mem_obj);

    ret = clReleaseCommandQueue(command_queue);

    ret = clReleaseContext(context);

free(C);

return 0;

}

You need to understand the relationship between pointers in private memory and data in local memory. The following line of code

__local int *a1 ;

declares a pointer a1 in private memory that points into local memory. You then make this pointer point to address 0 in local memory.

a1 = 0 ;

Nothing has touch the the actual local memory yet, you just changed the value of a variable in private memory. Then some work-item specific stuff happens when only work-item 1 sets its private instance of a1 to 100.

if ( t == 1)

   a1 = 100 ;

This is followed by a barrier…

barrier ( CLK_LOCAL_MEM_FENCE ) ;

…that will do nothing since no one has written anything to local memory yet. In the end you let work item 0 and 1 write their respective contents of the private variable a1 to global memory.

if ( t == 0 ) C [ t ] = a1 ;   

if ( t == 1 ) C [ t ] = a1 ;

These will be 0 and 100, respectively.

From your description of the problem I get the impression that what you really want is to have the variable a1 in local memory instead of private memory. Try changing the first line I quoted to

__local int a1 ;

I have a question myself. How do we declare a pointer in the local address space that points into the local address space? The rule so far has been to prepend the type name with __local, so by that rule it should be

__local __local int* a1 ;

Have never tried it though, and I don’t have any OpenCL compiler available at the moment. Does anyone know?